Hello readers,
Welcome to the AI For All newsletter! Today, we’re talking about AI taking work off scientists’ plates but adding back more, what agentic AI means for remote patient monitoring, and more!
AI in Action: AI is quietly sending scientists the bill
A new study from Google, Google DeepMind, and MIT FutureTech puts a number on what AI gives back to researchers: almost seven hours a week for the average scientist, close to a full workday. The team surveyed 637 scientists in the US and UK in July and August, then compared their answers with a sample of 15 million Gemini interactions and an inventory of more than 2,600 specialized models. About three-quarters reported saving time, and nearly half use AI every day. According to the paper, most of the recovered hours go back into research rather than shorter weeks.
The savings come with a bill. Nearly half of the scientists who reported a productivity gain say they spend 25% of the saved time checking what the AI produced, and the paper describes heavy demand for output verification. For a scientist saving seven hours, a quarter works out to roughly an hour and 45 minutes of review each week, a task that didn’t exist before. A model can draft the analysis or write the code in minutes. Confirming that either is correct still takes a person who knows the field well enough to catch what looks plausible but isn’t.
The other cost is a growing backlog of untested hypotheses. AI may have multiplied the number of viable ideas without adding any capacity to test them, and 44% of respondents say their main constraint has shifted downstream over the past two years, into lab work, clinical validation, and field data collection. AEI’s summary notes that experiments and data collection remain slow and labor-intensive. The authors caution that the survey may lean toward AI enthusiasts, who are likelier to respond. They also find that both the downstream shift and the backlog rise with how heavily a respondent uses AI.
🔥 Rapid Fire
Commentary: Credit Crunch
Analysis: How Has AI Changed The Economy?
OpenAI’s annualized revenue is $20 billion less than previously signaled
“OpenAI’s annualised revenue is about $20bn less than the company has previously signalled, according to financial documents shared with investors, a massive gap likely to damp optimism about the growth of AI demand.”
“The company has recently told investors its revenues were approaching $50bn on an annualised basis at the end of September, far short of the $70bn reported by the FT and other media outlets late last month based on information that was provided to investors.”
“US tech stocks fell sharply following the FT’s report on Thursday, extending losses from earlier in the session, with the Nasdaq 100 down 1.7 per cent. Chipmaker Nvidia fell 2.9 per cent, Oracle dropped almost 6 per cent and Micron declined 4 per cent.”
Microsoft and Meta cut spending on Anthropic’s Claude Code
“Microsoft reduced its projected internal spending on Claude by over one-third, while Meta saw internal users of Anthropic’s Claude Code coding assistant drop from roughly 60,000 to 30,000.”
“Rising costs and data privacy considerations are causing multiple enterprise clients to re-evaluate heavy reliance on Anthropic’s top-tier models ahead of its anticipated initial public offering.”
NVIDIA-backed Firmus misses data center payment and postpones IPO
Amazon wants to lease back $8 billion of its own chips through an SPV
Oracle, Broadcom, and SpaceX want to raise even more debt for AI chips
Spending on AI needs to rival food to justify investments
“To justify staggering investment sums, Americans will have to spend as much of their income on this one technology as they do on food.”
Meta rushed to fix Muse ‘VM Escape’ vulnerability soon before launch
📖 What We’re Reading
It’s 3 am, and the nurse on duty has already checked 50 pings in the remote monitoring dashboard from the evening. Most of these pings were insignificant clamor. The dashboard pings even when a reading drifted for a second or when a patient rolled over. But within that noise, there was one alert that needed immediate reporting to the doctor. Hours passed since that alert, and it has been overlooked in this pile of noise. Human attention in a demanding environment like a hospital often fails to catch the alert that matters from a pile of pure noise.
